@tangle-network/agent-runtime 0.89.0 → 0.90.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +51 -18
- package/dist/agent.d.ts +53 -2
- package/dist/agent.js +166 -4
- package/dist/agent.js.map +1 -1
- package/dist/analyst-loop.d.ts +1 -1
- package/dist/{chunk-PIPPLSOF.js → chunk-4IBAMGBE.js} +33 -3
- package/dist/chunk-4IBAMGBE.js.map +1 -0
- package/dist/chunk-5QOB7H74.js +387 -0
- package/dist/chunk-5QOB7H74.js.map +1 -0
- package/dist/chunk-74SBMDTO.js +214 -0
- package/dist/chunk-74SBMDTO.js.map +1 -0
- package/dist/{chunk-BQPFZE2C.js → chunk-MHK62APK.js} +5 -5
- package/dist/{chunk-N7EJV7N3.js → chunk-QSNSMJSZ.js} +3417 -3393
- package/dist/chunk-QSNSMJSZ.js.map +1 -0
- package/dist/{chunk-VNOOH22O.js → chunk-XMOU4WEY.js} +74 -3
- package/dist/chunk-XMOU4WEY.js.map +1 -0
- package/dist/{chunk-5AVV7KAH.js → chunk-XQEISTK2.js} +2 -2
- package/dist/{worktree-harness-CX_McRLp.d.ts → completion-gate-BDaBP8dd.d.ts} +44 -44
- package/dist/{coordination-CuDLO8wj.d.ts → coordination-DxHduZg7.d.ts} +414 -226
- package/dist/environment-provider.d.ts +3 -3
- package/dist/index.d.ts +156 -13
- package/dist/index.js +316 -70
- package/dist/index.js.map +1 -1
- package/dist/intelligence.d.ts +2 -2
- package/dist/knowledge.d.ts +120 -0
- package/dist/knowledge.js +28 -0
- package/dist/knowledge.js.map +1 -0
- package/dist/lifecycle.js +1 -1
- package/dist/{loop-runner-bin-hQDUL8Ld.d.ts → loop-runner-bin-D3GmB2Ru.d.ts} +2 -2
- package/dist/loop-runner-bin.d.ts +5 -5
- package/dist/loop-runner-bin.js +5 -5
- package/dist/loops.d.ts +46 -141
- package/dist/loops.js +3 -3
- package/dist/mcp/bin.js +2 -2
- package/dist/mcp/index.d.ts +8 -9
- package/dist/mcp/index.js +7 -7
- package/dist/{openai-tools-BnrOmGjN.d.ts → openai-tools-D0xmzo0y.d.ts} +1 -1
- package/dist/profiles.d.ts +196 -5
- package/dist/profiles.js +8 -0
- package/dist/profiles.js.map +1 -1
- package/dist/{router-client-r8y_VFVM.d.ts → router-client-DJImUDlm.d.ts} +1 -1
- package/dist/supervise-BpCdssu0.d.ts +131 -0
- package/dist/{types-Driepl87.d.ts → types-DAJQRIUD.d.ts} +2 -2
- package/dist/{types-ESeMOj94.d.ts → types-Dnk189QA.d.ts} +1 -1
- package/dist/{worktree-fanout-D6xR2CIA.d.ts → worktree-fanout-CfRXYmgV.d.ts} +3 -3
- package/package.json +12 -4
- package/skills/agent-runtime-adoption/SKILL.md +74 -24
- package/skills/build-with-agent-runtime/SKILL.md +3 -3
- package/skills/loop-writer/SKILL.md +6 -5
- package/skills/supervise/SKILL.md +2 -2
- package/dist/chunk-N7EJV7N3.js.map +0 -1
- package/dist/chunk-PIPPLSOF.js.map +0 -1
- package/dist/chunk-VNOOH22O.js.map +0 -1
- package/dist/otel-export-BKmNwiCb.d.ts +0 -180
- /package/dist/{chunk-BQPFZE2C.js.map → chunk-MHK62APK.js.map} +0 -0
- /package/dist/{chunk-5AVV7KAH.js.map → chunk-XQEISTK2.js.map} +0 -0
package/dist/intelligence.d.ts
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
|
-
import { c as LoopTraceEvent } from './types-
|
|
1
|
+
import { c as LoopTraceEvent } from './types-Dnk189QA.js';
|
|
2
2
|
import { AgentProfileMcpServer } from '@tangle-network/agent-interface';
|
|
3
|
-
import {
|
|
3
|
+
import { b as ToolSpec } from './router-client-DJImUDlm.js';
|
|
4
4
|
import '@tangle-network/agent-eval';
|
|
5
5
|
import '@tangle-network/sandbox';
|
|
6
6
|
|
|
@@ -0,0 +1,120 @@
|
|
|
1
|
+
import { KnowledgeReadinessSpec, BuildEvalKnowledgeBundleOptions, KnowledgeBaseQualityOptions, KnowledgeImprovementResult, KnowledgeImprovementOptions } from '@tangle-network/agent-knowledge';
|
|
2
|
+
import { E as ExecutorConfig } from './coordination-DxHduZg7.js';
|
|
3
|
+
import { a as SuperviseOptions, S as SupervisorProfile } from './supervise-BpCdssu0.js';
|
|
4
|
+
import { B as Budget, b as SupervisedResult } from './types-DAJQRIUD.js';
|
|
5
|
+
import { D as DeliverableSpec } from './completion-gate-BDaBP8dd.js';
|
|
6
|
+
import '@tangle-network/agent-interface';
|
|
7
|
+
import './substrate-DO2GHNg2.js';
|
|
8
|
+
import './types-Dnk189QA.js';
|
|
9
|
+
import '@tangle-network/agent-eval';
|
|
10
|
+
import '@tangle-network/sandbox';
|
|
11
|
+
import './router-client-DJImUDlm.js';
|
|
12
|
+
import './local-harness-sI0S_XNA.js';
|
|
13
|
+
import 'node:child_process';
|
|
14
|
+
import './environment-provider.js';
|
|
15
|
+
import '@tangle-network/agent-interface/environment-provider';
|
|
16
|
+
|
|
17
|
+
/** Standing prompt for a supervisor that grows a shared knowledge base through spawned researchers. */
|
|
18
|
+
declare const RESEARCH_SUPERVISOR_SYSTEM_PROMPT: string;
|
|
19
|
+
interface KnowledgeReadinessCheckInput {
|
|
20
|
+
root: string;
|
|
21
|
+
goal: string;
|
|
22
|
+
readinessSpecs?: readonly unknown[];
|
|
23
|
+
readinessTaskId?: string;
|
|
24
|
+
readiness?: unknown;
|
|
25
|
+
}
|
|
26
|
+
type KnowledgeReadinessCheckResult = boolean | {
|
|
27
|
+
ready: boolean;
|
|
28
|
+
summary?: string;
|
|
29
|
+
metadata?: Record<string, unknown>;
|
|
30
|
+
};
|
|
31
|
+
type KnowledgeReadinessCheck = (input: KnowledgeReadinessCheckInput) => Promise<KnowledgeReadinessCheckResult> | KnowledgeReadinessCheckResult;
|
|
32
|
+
interface SupervisedKnowledgeUpdateInput {
|
|
33
|
+
goal?: string;
|
|
34
|
+
root?: string;
|
|
35
|
+
candidateRoot?: string;
|
|
36
|
+
findings?: readonly unknown[];
|
|
37
|
+
metadata?: Record<string, unknown>;
|
|
38
|
+
}
|
|
39
|
+
interface SupervisedKnowledgeUpdateResult {
|
|
40
|
+
applied: boolean;
|
|
41
|
+
summary: string;
|
|
42
|
+
supervised: SupervisedResult<unknown>;
|
|
43
|
+
metadata: Record<string, unknown>;
|
|
44
|
+
}
|
|
45
|
+
interface SupervisedKnowledgeUpdateOptions {
|
|
46
|
+
root: string;
|
|
47
|
+
goal: string;
|
|
48
|
+
readiness: KnowledgeReadinessCheck;
|
|
49
|
+
readinessSpecs?: readonly unknown[];
|
|
50
|
+
readinessTaskId?: string;
|
|
51
|
+
readinessOptions?: unknown;
|
|
52
|
+
findings?: readonly unknown[];
|
|
53
|
+
metadata?: Record<string, unknown>;
|
|
54
|
+
budget: Budget;
|
|
55
|
+
backend?: ExecutorConfig;
|
|
56
|
+
makeWorkerAgent?: SuperviseOptions['makeWorkerAgent'];
|
|
57
|
+
harness?: string;
|
|
58
|
+
supervisorModel?: string;
|
|
59
|
+
supervisorSystemPrompt?: string;
|
|
60
|
+
superviseOptions?: Partial<Omit<SuperviseOptions, 'budget' | 'backend' | 'deliverable' | 'makeWorkerAgent' | 'allowedModels'>>;
|
|
61
|
+
allowedModels?: readonly string[];
|
|
62
|
+
runSupervised?: (profile: SupervisorProfile, task: unknown, opts: SuperviseOptions) => Promise<SupervisedResult<unknown>>;
|
|
63
|
+
}
|
|
64
|
+
type SupervisedKnowledgeUpdater = (input: SupervisedKnowledgeUpdateInput) => Promise<SupervisedKnowledgeUpdateResult>;
|
|
65
|
+
/** Build the completion check a supervised KB update uses to stop only when the KB is ready. */
|
|
66
|
+
declare function knowledgeReadinessDeliverable(options: Pick<SupervisedKnowledgeUpdateOptions, 'root' | 'goal' | 'readiness' | 'readinessSpecs' | 'readinessTaskId' | 'readinessOptions'>): DeliverableSpec<unknown>;
|
|
67
|
+
/** Create an `improveKnowledgeBase` update callback backed by runtime supervision. */
|
|
68
|
+
declare function createSupervisedKnowledgeUpdater(options: SupervisedKnowledgeUpdateOptions): SupervisedKnowledgeUpdater;
|
|
69
|
+
/** Run a runtime supervisor that updates one candidate knowledge base and stops on readiness. */
|
|
70
|
+
declare function runSupervisedKnowledgeUpdate(options: SupervisedKnowledgeUpdateOptions): Promise<SupervisedKnowledgeUpdateResult>;
|
|
71
|
+
/** Format the supervisor task with the KB root, readiness requirements, current findings, and metadata. */
|
|
72
|
+
declare function formatSupervisedKnowledgeTask(options: Pick<SupervisedKnowledgeUpdateOptions, 'root' | 'goal' | 'readinessSpecs' | 'readinessTaskId' | 'findings' | 'metadata'>): string;
|
|
73
|
+
|
|
74
|
+
interface RunKnowledgeImprovementJobOptions extends Omit<KnowledgeImprovementOptions, 'updateKnowledge'> {
|
|
75
|
+
budget: Budget;
|
|
76
|
+
readinessCheck?: KnowledgeReadinessCheck;
|
|
77
|
+
backend?: ExecutorConfig;
|
|
78
|
+
makeWorkerAgent?: SuperviseOptions['makeWorkerAgent'];
|
|
79
|
+
harness?: string;
|
|
80
|
+
supervisorModel?: string;
|
|
81
|
+
supervisorSystemPrompt?: string;
|
|
82
|
+
superviseOptions?: Partial<Omit<SuperviseOptions, 'budget' | 'backend' | 'deliverable' | 'makeWorkerAgent' | 'allowedModels'>>;
|
|
83
|
+
allowedModels?: readonly string[];
|
|
84
|
+
runSupervised?: (profile: SupervisorProfile, task: unknown, opts: SuperviseOptions) => Promise<SupervisedResult<unknown>>;
|
|
85
|
+
onMeasurement?: (measurement: KnowledgeImprovementJobMeasurement) => Promise<void> | void;
|
|
86
|
+
}
|
|
87
|
+
interface KnowledgeImprovementJobMeasurement {
|
|
88
|
+
startedAt: string;
|
|
89
|
+
finishedAt: string;
|
|
90
|
+
durationMs: number;
|
|
91
|
+
updateCalls: number;
|
|
92
|
+
updateDurationMs: number;
|
|
93
|
+
supervisedSpent: {
|
|
94
|
+
iterations: number;
|
|
95
|
+
inputTokens: number;
|
|
96
|
+
outputTokens: number;
|
|
97
|
+
usd: number;
|
|
98
|
+
ms: number;
|
|
99
|
+
};
|
|
100
|
+
}
|
|
101
|
+
interface KnowledgeImprovementJobResult {
|
|
102
|
+
improvement: KnowledgeImprovementResult;
|
|
103
|
+
measurement: KnowledgeImprovementJobMeasurement;
|
|
104
|
+
promoted: boolean;
|
|
105
|
+
blocked: boolean;
|
|
106
|
+
}
|
|
107
|
+
interface AgentKnowledgeReadinessCheckOptions {
|
|
108
|
+
goal: string;
|
|
109
|
+
readinessSpecs?: readonly KnowledgeReadinessSpec[];
|
|
110
|
+
readinessTaskId?: string;
|
|
111
|
+
readiness?: Omit<BuildEvalKnowledgeBundleOptions, 'taskId' | 'index' | 'specs'>;
|
|
112
|
+
strict?: boolean;
|
|
113
|
+
kbQuality?: KnowledgeBaseQualityOptions;
|
|
114
|
+
}
|
|
115
|
+
/** Build the default readiness check backed by `@tangle-network/agent-knowledge` validation and scoring. */
|
|
116
|
+
declare function createAgentKnowledgeReadinessCheck(options: AgentKnowledgeReadinessCheckOptions): KnowledgeReadinessCheck;
|
|
117
|
+
/** Run the full KB improvement job: candidate workspace, runtime supervisor update, readiness check, and promotion. */
|
|
118
|
+
declare function runKnowledgeImprovementJob(options: RunKnowledgeImprovementJobOptions): Promise<KnowledgeImprovementJobResult>;
|
|
119
|
+
|
|
120
|
+
export { type AgentKnowledgeReadinessCheckOptions, type KnowledgeImprovementJobMeasurement, type KnowledgeImprovementJobResult, type KnowledgeReadinessCheck, type KnowledgeReadinessCheckInput, type KnowledgeReadinessCheckResult, RESEARCH_SUPERVISOR_SYSTEM_PROMPT, type RunKnowledgeImprovementJobOptions, type SupervisedKnowledgeUpdateInput, type SupervisedKnowledgeUpdateOptions, type SupervisedKnowledgeUpdateResult, type SupervisedKnowledgeUpdater, createAgentKnowledgeReadinessCheck, createSupervisedKnowledgeUpdater, formatSupervisedKnowledgeTask, knowledgeReadinessDeliverable, runKnowledgeImprovementJob, runSupervisedKnowledgeUpdate };
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
import {
|
|
2
|
+
RESEARCH_SUPERVISOR_SYSTEM_PROMPT,
|
|
3
|
+
createAgentKnowledgeReadinessCheck,
|
|
4
|
+
createSupervisedKnowledgeUpdater,
|
|
5
|
+
formatSupervisedKnowledgeTask,
|
|
6
|
+
knowledgeReadinessDeliverable,
|
|
7
|
+
runKnowledgeImprovementJob,
|
|
8
|
+
runSupervisedKnowledgeUpdate
|
|
9
|
+
} from "./chunk-74SBMDTO.js";
|
|
10
|
+
import "./chunk-5QOB7H74.js";
|
|
11
|
+
import "./chunk-QSNSMJSZ.js";
|
|
12
|
+
import "./chunk-DPEUKJRO.js";
|
|
13
|
+
import "./chunk-UD4BHQMI.js";
|
|
14
|
+
import "./chunk-BZF3KQ6G.js";
|
|
15
|
+
import "./chunk-FVJ7M3DA.js";
|
|
16
|
+
import "./chunk-7LO5GMAO.js";
|
|
17
|
+
import "./chunk-YEJR7IXO.js";
|
|
18
|
+
import "./chunk-DGUM43GV.js";
|
|
19
|
+
export {
|
|
20
|
+
RESEARCH_SUPERVISOR_SYSTEM_PROMPT,
|
|
21
|
+
createAgentKnowledgeReadinessCheck,
|
|
22
|
+
createSupervisedKnowledgeUpdater,
|
|
23
|
+
formatSupervisedKnowledgeTask,
|
|
24
|
+
knowledgeReadinessDeliverable,
|
|
25
|
+
runKnowledgeImprovementJob,
|
|
26
|
+
runSupervisedKnowledgeUpdate
|
|
27
|
+
};
|
|
28
|
+
//# sourceMappingURL=knowledge.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}
|
package/dist/lifecycle.js
CHANGED
|
@@ -2,8 +2,8 @@ import { Scenario } from '@tangle-network/agent-eval/campaign';
|
|
|
2
2
|
import { SelfImproveOptions, SelfImproveResult } from '@tangle-network/agent-eval/contract';
|
|
3
3
|
import { R as RunAnalystLoopOpts, a as RunAnalystLoopResult } from './types-BC3bZpH0.js';
|
|
4
4
|
import { F as FactCandidate, C as CreateKbGateOptions } from './kb-gate-CwHO0vz6.js';
|
|
5
|
-
import { B as Budget } from './types-
|
|
6
|
-
import { A as AuthoredHarness, W as WinnerStrategy, a as WorktreeFanoutOptions, b as WorktreePatchArtifact } from './worktree-fanout-
|
|
5
|
+
import { B as Budget } from './types-DAJQRIUD.js';
|
|
6
|
+
import { A as AuthoredHarness, W as WinnerStrategy, a as WorktreeFanoutOptions, b as WorktreePatchArtifact } from './worktree-fanout-CfRXYmgV.js';
|
|
7
7
|
|
|
8
8
|
/**
|
|
9
9
|
*
|
|
@@ -1,15 +1,15 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
|
-
export { L as LoopRunnerCliArgs, e as LoopRunnerCliResult, p as parseLoopRunnerArgv, k as runLoopRunnerCli } from './loop-runner-bin-
|
|
2
|
+
export { L as LoopRunnerCliArgs, e as LoopRunnerCliResult, p as parseLoopRunnerArgv, k as runLoopRunnerCli } from './loop-runner-bin-D3GmB2Ru.js';
|
|
3
3
|
import '@tangle-network/agent-eval/campaign';
|
|
4
4
|
import '@tangle-network/agent-eval/contract';
|
|
5
5
|
import './types-BC3bZpH0.js';
|
|
6
6
|
import '@tangle-network/agent-eval';
|
|
7
7
|
import './kb-gate-CwHO0vz6.js';
|
|
8
|
-
import './types-
|
|
8
|
+
import './types-DAJQRIUD.js';
|
|
9
9
|
import '@tangle-network/agent-interface';
|
|
10
10
|
import '@tangle-network/sandbox';
|
|
11
|
-
import './types-
|
|
12
|
-
import './worktree-fanout-
|
|
13
|
-
import './
|
|
11
|
+
import './types-Dnk189QA.js';
|
|
12
|
+
import './worktree-fanout-CfRXYmgV.js';
|
|
13
|
+
import './completion-gate-BDaBP8dd.js';
|
|
14
14
|
import './local-harness-sI0S_XNA.js';
|
|
15
15
|
import 'node:child_process';
|
package/dist/loop-runner-bin.js
CHANGED
|
@@ -2,17 +2,17 @@
|
|
|
2
2
|
import {
|
|
3
3
|
parseLoopRunnerArgv,
|
|
4
4
|
runLoopRunnerCli
|
|
5
|
-
} from "./chunk-
|
|
5
|
+
} from "./chunk-MHK62APK.js";
|
|
6
6
|
import "./chunk-SGKPNBXE.js";
|
|
7
|
-
import "./chunk-
|
|
8
|
-
import "./chunk-
|
|
7
|
+
import "./chunk-4IBAMGBE.js";
|
|
8
|
+
import "./chunk-ZQZX77MM.js";
|
|
9
|
+
import "./chunk-QSNSMJSZ.js";
|
|
10
|
+
import "./chunk-DPEUKJRO.js";
|
|
9
11
|
import "./chunk-UD4BHQMI.js";
|
|
10
12
|
import "./chunk-BZF3KQ6G.js";
|
|
11
|
-
import "./chunk-ZQZX77MM.js";
|
|
12
13
|
import "./chunk-FVJ7M3DA.js";
|
|
13
14
|
import "./chunk-7LO5GMAO.js";
|
|
14
15
|
import "./chunk-YEJR7IXO.js";
|
|
15
|
-
import "./chunk-DPEUKJRO.js";
|
|
16
16
|
import "./chunk-DGUM43GV.js";
|
|
17
17
|
export {
|
|
18
18
|
parseLoopRunnerArgv,
|
package/dist/loops.d.ts
CHANGED
|
@@ -2,26 +2,27 @@ import { ChatClient, RunRecord, HarnessType, AgentProfile, AnalystFinding, Analy
|
|
|
2
2
|
export { AnalystFinding, DefaultVerdict, computeFindingId, makeFinding } from '@tangle-network/agent-eval';
|
|
3
3
|
import { SandboxEvent, SandboxInstance, CreateSandboxOptions, PromptOptions, TaskOptions } from '@tangle-network/sandbox';
|
|
4
4
|
export { AgentProfile, CreateSandboxOptions, SandboxEvent, SandboxInstance } from '@tangle-network/sandbox';
|
|
5
|
-
import {
|
|
6
|
-
export {
|
|
7
|
-
import {
|
|
8
|
-
export {
|
|
9
|
-
import { R as RuntimeHooks, I as Iteration, S as SandboxClient,
|
|
10
|
-
export { $ as LoopDecisionPayload, a0 as LoopEndedPayload, a1 as LoopIterationDispatchPayload, a2 as LoopIterationEndedPayload, a3 as LoopIterationStartedPayload, a4 as LoopPlanDescription, a5 as LoopPlanPayload, d as LoopSandboxPlacement, a6 as LoopStartedPayload, a7 as LoopTeardownFailedPayload, e as LoopTraceEmitter, c as LoopTraceEvent, a8 as MountManifestEntry, a9 as RunProvenance, aa as SelectionReceipt, ab as ValidationCtx } from './types-
|
|
5
|
+
import { a as ResultBlobStore, f as SpawnJournal, N as NodeId, j as SpawnEvent, E as ExecutorFactory, A as Agent, B as Budget, S as Scope, h as Settled, b as SupervisedResult, i as Spend, U as UsageEvent, e as ExecutorRegistry, k as Supervisor } from './types-DAJQRIUD.js';
|
|
6
|
+
export { d as AgentSpec, c as Executor, l as ExecutorContext, m as ExecutorResult, R as Runtime, n as SupervisorOpts, T as TreeView, W as WidenGate } from './types-DAJQRIUD.js';
|
|
7
|
+
import { M as MakeWorkerAgent, A as AnalystRegistry, o as CoordinationTools, n as CoordinationEvent, as as QuestionPolicy, E as ExecutorConfig } from './coordination-DxHduZg7.js';
|
|
8
|
+
export { bf as BusEvent, bg as BusRecord, bh as BusStats, bi as EventBus, bj as ProviderSeam, bk as PublishOptions, bl as cliWorktreeExecutor, bm as createEventBus, bn as createExecutor, bo as createExecutorRegistry } from './coordination-DxHduZg7.js';
|
|
9
|
+
import { R as RuntimeHooks, I as Iteration, S as SandboxClient, D as Driver, A as AgentRunSpec, b as OutputAdapter, V as Validator, E as ExecCtx, X as LoopWinner, Y as LoopLineageOptions, Z as LoopResult, L as LoopTokenUsage, a as RuntimeStreamEvent, _ as MountRecorder, i as AgentExecutionBackend, o as AgentTaskStatus, B as BackendErrorDetail } from './types-Dnk189QA.js';
|
|
10
|
+
export { $ as LoopDecisionPayload, a0 as LoopEndedPayload, a1 as LoopIterationDispatchPayload, a2 as LoopIterationEndedPayload, a3 as LoopIterationStartedPayload, a4 as LoopPlanDescription, a5 as LoopPlanPayload, d as LoopSandboxPlacement, a6 as LoopStartedPayload, a7 as LoopTeardownFailedPayload, e as LoopTraceEmitter, c as LoopTraceEvent, a8 as MountManifestEntry, a9 as RunProvenance, aa as SelectionReceipt, ab as ValidationCtx } from './types-Dnk189QA.js';
|
|
11
11
|
import { RunProfileMatrixResult, Scenario, ProfileDispatchFn, JudgeConfig, RunProfileMatrixOptions, DispatchFn } from '@tangle-network/agent-eval/campaign';
|
|
12
12
|
export { AgentEnvironmentProviderRef, AgentEnvironmentProviderRegistry, ProviderAsSandboxClientOptions, ProviderExecutorOptions, SandboxClientProviderOptions, createAgentEnvironmentProviderRegistry, providerAsExecutor, providerAsSandboxClient, resolveAgentEnvironmentProvider, sandboxClientAsProvider } from './environment-provider.js';
|
|
13
|
-
import { C as CorpusRecord, c as Corpus, O as Outcome, S as ScopeAnalyzeInput, d as AssertTraceDerivedFindings, e as SteerContext, f as ScopeAnalyst, F as FanoutOptions, g as CombinatorShape, h as ScopeWidenGate, L as LoopUntilSpec, P as PanelSpec, i as PipelineStage, W as WinnerStrategy, j as FanoutWinnerSelector, V as VerifySpec, k as WidenSpec, l as CorpusFilter, R as RenderCorpusToInstructionsOptions, D as DefinePersonaInput, m as Persona, n as RunPersonifiedOptions, o as ShapeRegistry, p as LoopShape, E as EqualKArm, q as EqualKOnCostOptions, r as EqualKVerdict, T as TrajectoryReportOptions, s as TrajectoryReport } from './worktree-fanout-
|
|
14
|
-
export { A as AuthoredHarness, t as DefinePersona, u as EqualKOnCost, v as Fanout, w as FanoutSynthesis, x as FlatWidenGate, y as LoopUntil, z as LoopUntilState, B as Panel, G as PanelJudge, H as PanelVerdict, I as PatchDeliverableOptions, J as PersonaContext, K as PersonaExecutors, M as Pipeline, N as RenderCorpusToInstructions, Q as RunPersonified, U as ShapeBudget, X as ShapeContext, Y as TrajectoryNode, Z as TrajectoryReportFn, _ as Verify, $ as Widen, a0 as WidenDecision, a1 as WidenLineage, a2 as WorktreeCliExecutorOptions, a as WorktreeFanoutOptions, b as WorktreePatchArtifact, a3 as createWorktreeCliExecutor, a4 as patchDelivered, a5 as worktreeFanout } from './worktree-fanout-
|
|
13
|
+
import { C as CorpusRecord, c as Corpus, O as Outcome, S as ScopeAnalyzeInput, d as AssertTraceDerivedFindings, e as SteerContext, f as ScopeAnalyst, F as FanoutOptions, g as CombinatorShape, h as ScopeWidenGate, L as LoopUntilSpec, P as PanelSpec, i as PipelineStage, W as WinnerStrategy, j as FanoutWinnerSelector, V as VerifySpec, k as WidenSpec, l as CorpusFilter, R as RenderCorpusToInstructionsOptions, D as DefinePersonaInput, m as Persona, n as RunPersonifiedOptions, o as ShapeRegistry, p as LoopShape, E as EqualKArm, q as EqualKOnCostOptions, r as EqualKVerdict, T as TrajectoryReportOptions, s as TrajectoryReport } from './worktree-fanout-CfRXYmgV.js';
|
|
14
|
+
export { A as AuthoredHarness, t as DefinePersona, u as EqualKOnCost, v as Fanout, w as FanoutSynthesis, x as FlatWidenGate, y as LoopUntil, z as LoopUntilState, B as Panel, G as PanelJudge, H as PanelVerdict, I as PatchDeliverableOptions, J as PersonaContext, K as PersonaExecutors, M as Pipeline, N as RenderCorpusToInstructions, Q as RunPersonified, U as ShapeBudget, X as ShapeContext, Y as TrajectoryNode, Z as TrajectoryReportFn, _ as Verify, $ as Widen, a0 as WidenDecision, a1 as WidenLineage, a2 as WorktreeCliExecutorOptions, a as WorktreeFanoutOptions, b as WorktreePatchArtifact, a3 as createWorktreeCliExecutor, a4 as patchDelivered, a5 as worktreeFanout } from './worktree-fanout-CfRXYmgV.js';
|
|
15
15
|
import { b as AnalystRegistryLike } from './types-BC3bZpH0.js';
|
|
16
16
|
import { AgentProfile as AgentProfile$1 } from '@tangle-network/agent-interface';
|
|
17
|
-
import { R as RouterConfig,
|
|
18
|
-
export { c as RouterChatResult, d as RouterChatToolsResult, e as RouterToolCall, f as RouterToolLoopResult, g as ToolLoopCompaction,
|
|
19
|
-
import { D as DeliverableSpec } from './
|
|
20
|
-
export { g as WorktreeCommandResult, h as gateOnDeliverable } from './
|
|
17
|
+
import { R as RouterConfig, T as ToolLoopChat, a as ToolLoopCompactionOptions } from './router-client-DJImUDlm.js';
|
|
18
|
+
export { c as RouterChatResult, d as RouterChatToolsResult, e as RouterToolCall, f as RouterToolLoopResult, g as ToolLoopCompaction, b as ToolSpec, r as routerBrain, h as routerChatWithTools, i as routerChatWithUsage, j as routerToolLoop } from './router-client-DJImUDlm.js';
|
|
19
|
+
import { D as DeliverableSpec } from './completion-gate-BDaBP8dd.js';
|
|
20
|
+
export { g as WorktreeCommandResult, h as gateOnDeliverable } from './completion-gate-BDaBP8dd.js';
|
|
21
|
+
import { S as SupervisorProfile } from './supervise-BpCdssu0.js';
|
|
22
|
+
export { D as DriveHarness, a as SuperviseOptions, b as SupervisorAgentDeps, s as supervise, c as supervisorAgent, w as workerFromBackend } from './supervise-BpCdssu0.js';
|
|
21
23
|
import { stuckLoopView, toolWasteView } from '@tangle-network/agent-eval/pipelines';
|
|
22
24
|
export { AgentEnvironment, AgentEnvironmentCapabilities, AgentEnvironmentEvent, AgentEnvironmentProvider, AgentEnvironmentQuery, AgentEnvironmentStatus, AgentEnvironmentSummary, AgentProfileRef, AgentSession, AgentSessionRef, AgentSessionStatus, AgentTurnInput, AgentTurnResult, CheckpointRef, CheckpointRequest, CreateAgentEnvironmentInput, ExecRequest, ExecResult, ForkRequest, PlacementInfo, ResourceRequest, WorkspaceRequest } from '@tangle-network/agent-interface/environment-provider';
|
|
23
25
|
import './substrate-DO2GHNg2.js';
|
|
24
|
-
import './otel-export-BKmNwiCb.js';
|
|
25
26
|
import './local-harness-sI0S_XNA.js';
|
|
26
27
|
import 'node:child_process';
|
|
27
28
|
|
|
@@ -1141,11 +1142,24 @@ interface AgenticOptions {
|
|
|
1141
1142
|
* worker. Omitted ⇒ the worker's `model`. */
|
|
1142
1143
|
analystModel?: string;
|
|
1143
1144
|
/** Across-run learning: when set, the analyst's observe() pass appends trace-derived
|
|
1144
|
-
* facts here (the flywheel write side).
|
|
1145
|
-
*
|
|
1145
|
+
* facts here (the flywheel write side). Read-back is opt-in via `corpusReadback`
|
|
1146
|
+
* because unconditional priming can pollute context on some domains. */
|
|
1146
1147
|
corpus?: Corpus;
|
|
1147
1148
|
/** Tags written onto learned facts (and used by the caller's priming query). */
|
|
1148
1149
|
corpusTags?: string[];
|
|
1150
|
+
/** In-context learning: when set, query `corpus` before each depth shot and inject
|
|
1151
|
+
* the top trace-derived facts as guidance for the active run. No corpus means no read-back. */
|
|
1152
|
+
corpusReadback?: CorpusReadbackOptions;
|
|
1153
|
+
}
|
|
1154
|
+
interface CorpusReadbackOptions {
|
|
1155
|
+
/** Minimum confidence for a fact to be injected. Default 0.7. */
|
|
1156
|
+
minConfidence?: number;
|
|
1157
|
+
/** Extra tags a fact must carry, in addition to `corpusTags`. */
|
|
1158
|
+
tags?: ReadonlyArray<string>;
|
|
1159
|
+
/** Max facts injected per shot. Default 3. */
|
|
1160
|
+
maxFacts?: number;
|
|
1161
|
+
/** Default false: only facts tagged `audience:agent` are injected into the worker. */
|
|
1162
|
+
includeOperatorFacts?: boolean;
|
|
1149
1163
|
}
|
|
1150
1164
|
type Msg = Record<string, unknown>;
|
|
1151
1165
|
interface ShotResult {
|
|
@@ -2432,6 +2446,14 @@ interface SandboxRun<Out> {
|
|
|
2432
2446
|
* @experimental
|
|
2433
2447
|
*/
|
|
2434
2448
|
type OpenSandboxRunPromptOptions = Omit<PromptOptions, 'signal' | 'sessionId'>;
|
|
2449
|
+
/** Context available after the box/session exists and before the first prompt is
|
|
2450
|
+
* drained. Intended for benchmark-owned workspace setup such as cloning a repo
|
|
2451
|
+
* into a fixed path. */
|
|
2452
|
+
interface OpenSandboxRunBeforeStartContext {
|
|
2453
|
+
readonly box: SandboxInstance;
|
|
2454
|
+
readonly sessionId: string;
|
|
2455
|
+
readonly signal: AbortSignal;
|
|
2456
|
+
}
|
|
2435
2457
|
/** @experimental */
|
|
2436
2458
|
interface OpenSandboxRunOptions {
|
|
2437
2459
|
/** Profile + sandbox env/overrides. `sandboxOverrides.backend.type` is the harness. */
|
|
@@ -2446,6 +2468,10 @@ interface OpenSandboxRunOptions {
|
|
|
2446
2468
|
/** Per-prompt sandbox SDK options forwarded to both `start()` and `resume()`.
|
|
2447
2469
|
* The runtime still owns the session id and abort signal for each turn. */
|
|
2448
2470
|
promptOptions?: OpenSandboxRunPromptOptions;
|
|
2471
|
+
/** Optional pre-start workspace setup. Runs after `lineage.start()` creates the
|
|
2472
|
+
* box/session and before the first prompt stream is consumed. A thrown error
|
|
2473
|
+
* fails the turn before the agent spends tokens. */
|
|
2474
|
+
beforeStart?: (ctx: OpenSandboxRunBeforeStartContext) => Promise<void> | void;
|
|
2449
2475
|
/** Test seam for deterministic hook timestamps. Defaults to `Date.now`. */
|
|
2450
2476
|
now?: () => number;
|
|
2451
2477
|
/** Bounds box-creation bursts inside lineage fanout. Default from lineage. */
|
|
@@ -3392,6 +3418,10 @@ declare function serveCoordinationMcp(opts: {
|
|
|
3392
3418
|
/** Hard cap on simultaneously-LIVE workers — `spawn_agent` fails closed once this many are in
|
|
3393
3419
|
* flight (a concurrency fence on top of the conserved-pool fence). Omit/`<= 0` = no cap. */
|
|
3394
3420
|
maxLiveWorkers?: number;
|
|
3421
|
+
/** Max wall-clock ms a single `await_event` may block before returning a re-pollable
|
|
3422
|
+
* `{ pending, live }` snapshot instead of erroring on the client's request timeout. Omit =
|
|
3423
|
+
* {@link DEFAULT_AWAIT_EVENT_TIMEOUT_MS}; `<= 0` = prior unbounded block (in-process only). */
|
|
3424
|
+
awaitTimeoutMs?: number;
|
|
3395
3425
|
port?: number;
|
|
3396
3426
|
host?: string;
|
|
3397
3427
|
/** Trace-analyst lenses the driver can run (`run_analyst`) or auto-fire on settle. */
|
|
@@ -3403,66 +3433,6 @@ declare function serveCoordinationMcp(opts: {
|
|
|
3403
3433
|
questionPolicy?: QuestionPolicy;
|
|
3404
3434
|
}): Promise<CoordinationMcpHandle>;
|
|
3405
3435
|
|
|
3406
|
-
/** The supervisor's profile — the subset of an `AgentProfile` that selects + shapes its brain.
|
|
3407
|
-
* `harness` is the backend-as-data discriminant; `systemPrompt` is the standing instruction. */
|
|
3408
|
-
interface SupervisorProfile {
|
|
3409
|
-
readonly name?: string;
|
|
3410
|
-
/** null/undefined → router brain (in-process tool-loop); a coding-CLI harness → sandboxed brain. */
|
|
3411
|
-
readonly harness?: string | null;
|
|
3412
|
-
/** The router model when the brain is router-driven (falls back to the deps router config). */
|
|
3413
|
-
readonly model?: string;
|
|
3414
|
-
/** The standing instructions ("you delegate, you do not solve"). */
|
|
3415
|
-
readonly systemPrompt?: string;
|
|
3416
|
-
}
|
|
3417
|
-
/** How to run a sandboxed harness as the DRIVER, with the coordination verbs mounted — the substrate
|
|
3418
|
-
* seam the caller supplies (mirrors `makeWorkerAgent` for spawned children). It runs `profile` on
|
|
3419
|
-
* `task` in its backend (sandbox / cli-bridge) with `coordinationMcpUrl` mounted as an MCP server,
|
|
3420
|
-
* so the harness calls spawn_agent / await_event / stop as native tools over the live scope. */
|
|
3421
|
-
type DriveHarness = (args: {
|
|
3422
|
-
readonly profile: SupervisorProfile;
|
|
3423
|
-
readonly task: unknown;
|
|
3424
|
-
readonly scope: Scope<unknown>;
|
|
3425
|
-
readonly coordinationMcpUrl: string;
|
|
3426
|
-
}) => Promise<void>;
|
|
3427
|
-
interface SupervisorAgentDeps {
|
|
3428
|
-
readonly blobs: ResultBlobStore;
|
|
3429
|
-
/** Resolve a spawned worker `profile` to a leaf agent — the recursion seam (same for both arms). */
|
|
3430
|
-
readonly makeWorkerAgent: MakeWorkerAgent;
|
|
3431
|
-
/** Per-child budget reserved from the conserved pool on each spawn. */
|
|
3432
|
-
readonly perWorker: Budget;
|
|
3433
|
-
/** Hard cap on simultaneously-LIVE workers across both arms — `spawn_agent` fails closed once
|
|
3434
|
-
* this many are in flight (a concurrency fence on top of the conserved-pool fence; bounds live
|
|
3435
|
-
* boxes/sandboxes, not total work). Omit/`<= 0` = no cap. */
|
|
3436
|
-
readonly maxLiveWorkers?: number;
|
|
3437
|
-
/** Router substrate for a router-brained supervisor (`harness` null). The profile's model wins. */
|
|
3438
|
-
readonly router?: RouterConfig;
|
|
3439
|
-
/** Inject the brain directly (tests / advanced) instead of resolving `routerBrain` from the profile. */
|
|
3440
|
-
readonly brain?: ToolLoopChat;
|
|
3441
|
-
/** Required for a sandboxed-harness supervisor (`harness` set): runs the harness as the driver. */
|
|
3442
|
-
readonly driveHarness?: DriveHarness;
|
|
3443
|
-
/** WORK tools the supervisor may call DIRECTLY (router arm) — so it can do simple work ITSELF and
|
|
3444
|
-
* only delegate when it needs parallelism. Pair with `executeExtraTool`. */
|
|
3445
|
-
readonly extraTools?: ReadonlyArray<{
|
|
3446
|
-
readonly name: string;
|
|
3447
|
-
readonly description?: string;
|
|
3448
|
-
readonly parameters: Record<string, unknown>;
|
|
3449
|
-
}>;
|
|
3450
|
-
/** Runs an `extraTools` call; null/undefined falls through to the coordination dispatch. */
|
|
3451
|
-
readonly executeExtraTool?: (name: string, args: Record<string, unknown>) => Promise<string | null | undefined>;
|
|
3452
|
-
/** Analyst lenses available to the driver (both arms). Required for `analyzeOnSettle`. */
|
|
3453
|
-
readonly analysts?: AnalystRegistry;
|
|
3454
|
-
/** Analyst kinds run on each worker-settle → a `finding` the driver composes its next steer from
|
|
3455
|
-
* (the self-improving UP-leg). Unset/empty = status quo (no analyst feed). Requires `analysts`. */
|
|
3456
|
-
readonly analyzeOnSettle?: ReadonlyArray<string>;
|
|
3457
|
-
readonly maxTurns?: number;
|
|
3458
|
-
/** Give the supervisor brain a chapter-lifecycle on its OWN context window (router arm only) — it
|
|
3459
|
-
* distills its coordination transcript to a compact progress note once it exceeds the threshold,
|
|
3460
|
-
* instead of re-billing the whole thing every turn. See `DriverAgentOptions.compaction`. */
|
|
3461
|
-
readonly compaction?: ToolLoopCompactionOptions;
|
|
3462
|
-
}
|
|
3463
|
-
/** Build a supervisor `Agent` from its profile: the brain resolves from `profile.harness` (backend-as-data), the same resolution rule as every worker. */
|
|
3464
|
-
declare function supervisorAgent(profile: SupervisorProfile, deps: SupervisorAgentDeps): Agent<unknown, unknown>;
|
|
3465
|
-
|
|
3466
3436
|
/**
|
|
3467
3437
|
*
|
|
3468
3438
|
* `delegate` — the one generic delegation verb. You hand it an INTENT (what you want done) and it
|
|
@@ -3803,71 +3773,6 @@ declare function createScope<Out>(args: ScopeArgs): Scope<Out>;
|
|
|
3803
3773
|
*/
|
|
3804
3774
|
declare function settledToIteration<Out>(settled: Settled<Out>): Iteration<unknown, Out>;
|
|
3805
3775
|
|
|
3806
|
-
/** Build the worker seam from a backend (WHERE workers run) + an optional completion oracle (the
|
|
3807
|
-
* deliverable check that makes "settled ⟺ delivered" true — the guard against "ran but didn't
|
|
3808
|
-
* deliver"). The ONE place a backend becomes a spawnable worker. */
|
|
3809
|
-
declare function workerFromBackend(backend: ExecutorConfig, deliverable?: DeliverableSpec<unknown>): MakeWorkerAgent;
|
|
3810
|
-
interface SuperviseOptions {
|
|
3811
|
-
/** The conserved compute pool for the whole run. */
|
|
3812
|
-
readonly budget: Budget;
|
|
3813
|
-
/** WHERE workers run — derives the worker seam. Provide this OR an explicit `makeWorkerAgent`. */
|
|
3814
|
-
readonly backend?: ExecutorConfig;
|
|
3815
|
-
/** The completion oracle for backend-derived workers (settled ⟺ delivered). Strongly recommended:
|
|
3816
|
-
* without it the supervisor trusts a worker's self-report — exactly the "ran but didn't deliver"
|
|
3817
|
-
* failure mode of a static orchestrator. */
|
|
3818
|
-
readonly deliverable?: DeliverableSpec<unknown>;
|
|
3819
|
-
/** Override the worker seam directly (tests / advanced) instead of deriving it from `backend`. */
|
|
3820
|
-
readonly makeWorkerAgent?: MakeWorkerAgent;
|
|
3821
|
-
/** The supervisor's router substrate (`harness` null). The profile's model wins. */
|
|
3822
|
-
readonly router?: RouterConfig;
|
|
3823
|
-
/** Inject the supervisor brain directly (tests / advanced). */
|
|
3824
|
-
readonly brain?: ToolLoopChat;
|
|
3825
|
-
/** Run a sandboxed-harness supervisor (`harness` set). */
|
|
3826
|
-
readonly driveHarness?: DriveHarness;
|
|
3827
|
-
/** WORK tools the supervisor may call DIRECTLY — so a recursive atom can ACT (do simple work
|
|
3828
|
-
* itself) OR SPAWN (delegate when it needs parallelism), not be a pure manager. Pair with
|
|
3829
|
-
* `executeExtraTool`. Router arm only (`harness` null). */
|
|
3830
|
-
readonly extraTools?: ReadonlyArray<{
|
|
3831
|
-
readonly name: string;
|
|
3832
|
-
readonly description?: string;
|
|
3833
|
-
readonly parameters: Record<string, unknown>;
|
|
3834
|
-
}>;
|
|
3835
|
-
/** Runs an `extraTools` call; null/undefined falls through to the coordination dispatch. */
|
|
3836
|
-
readonly executeExtraTool?: (name: string, args: Record<string, unknown>) => Promise<string | null | undefined>;
|
|
3837
|
-
/** Per-child budget reserved on each spawn. Defaults to a quarter of the pool's tokens. */
|
|
3838
|
-
readonly perWorker?: Budget;
|
|
3839
|
-
/** Hard cap on simultaneously-LIVE workers — `spawn_agent` fails closed once this many are in
|
|
3840
|
-
* flight. The conserved pool bounds TOTAL work; this bounds SIMULTANEOUS work (live boxes/
|
|
3841
|
-
* sandboxes a real fleet runs at once). Omit/`<= 0` = no cap (the pool stays the only fence). */
|
|
3842
|
-
readonly maxLiveWorkers?: number;
|
|
3843
|
-
/** Analyst lenses available to the driver. Required for `analyzeOnSettle`. Unset → status quo
|
|
3844
|
-
* (the driver receives settled worker outputs, no analyst findings). */
|
|
3845
|
-
readonly analysts?: AnalystRegistry;
|
|
3846
|
-
/** Analyst kind ids run AUTOMATICALLY when a worker settles `done` — each re-enters as a `finding`
|
|
3847
|
-
* the driver pulls (`await_event`) and composes its next steer from. The self-improving UP-leg,
|
|
3848
|
-
* threaded to the driver at this level (propagate to sub-drivers via a recursive `makeWorkerAgent`).
|
|
3849
|
-
* Omit/empty = status quo (no analyst feed). Requires `analysts`. */
|
|
3850
|
-
readonly analyzeOnSettle?: ReadonlyArray<string>;
|
|
3851
|
-
/** Worker output store. Defaults to in-memory. */
|
|
3852
|
-
readonly blobs?: ResultBlobStore;
|
|
3853
|
-
readonly maxDepth?: number;
|
|
3854
|
-
readonly maxTurns?: number;
|
|
3855
|
-
/** Give the supervisor brain a chapter-lifecycle on its OWN context window (router arm only): once
|
|
3856
|
-
* its coordination transcript exceeds `thresholdTokens` it distills to a compact progress note and
|
|
3857
|
-
* continues, instead of re-billing the whole transcript every turn (the cost that makes the LLM-brain
|
|
3858
|
-
* front door lose to a dumb-Ralph respawn). The live `Scope` roster is the durable state across
|
|
3859
|
-
* chapters. Default off. `distill` defaults to a brain self-summary + the settled-worker roster. */
|
|
3860
|
-
readonly compaction?: ToolLoopCompactionOptions;
|
|
3861
|
-
readonly runId?: string;
|
|
3862
|
-
readonly now?: () => number;
|
|
3863
|
-
/** Restrict the run to this subset of models. When set, every configured model — the
|
|
3864
|
-
* supervisor router model, the profile's model, and the backend's model — must be a member,
|
|
3865
|
-
* or `supervise()` throws a `ConfigError` before any compute is spent. Unset = unrestricted. */
|
|
3866
|
-
readonly allowedModels?: readonly string[];
|
|
3867
|
-
}
|
|
3868
|
-
/** One-call supervisor: build + run a supervisor from its profile with sensible defaults; the raw `supervisorAgent` + `createSupervisor().run` seams stay available for power use. */
|
|
3869
|
-
declare function supervise(profile: SupervisorProfile, task: unknown, opts: SuperviseOptions): Promise<SupervisedResult<unknown>>;
|
|
3870
|
-
|
|
3871
3776
|
/**
|
|
3872
3777
|
*
|
|
3873
3778
|
* The `Supervisor` impl (KEYSTONE, build step 5).
|
|
@@ -4089,4 +3994,4 @@ declare function runInWorkspace<T>(ws: Workspace, body: (cwd: string) => Promise
|
|
|
4089
3994
|
commitOnInvalid?: boolean;
|
|
4090
3995
|
}): Promise<WorkspaceRun<T>>;
|
|
4091
3996
|
|
|
4092
|
-
export { Agent, AgentRunSpec, type AgentTurnBackend, type AgentTurnUsage, type AgenticOptions, type AgenticRunResult, type AgenticSurface, type AgenticTask, type AgenticTool, AnalystRegistry, type AnytimeReport, type AnytimeStrategySummary, type AnytimeTaskCurve, type ApplyContinuation, type ArtifactHandle, AssertTraceDerivedFindings, type AuditIntentInput, type AuditIntentOptions, type AuthorStrategyOptions, type AuthoredProfile, type AuthoredStrategy, type BenchmarkCell, type BenchmarkConfig, type BenchmarkLift, type BenchmarkReport, type BenchmarkStrategySummary, type BenchmarkTaskRow, Budget, type BudgetPool, type BudgetReadout, type ChampionPick, type ChampionPolicy, type CheckpointCapableBox, type CollectedAgentTurn, CombinatorShape, type CompletionAnalyst, type CompletionEvidence, type CompletionPolicy, type CompletionVerdict, CoordinationEvent, type CoordinationMcpHandle, Corpus, CorpusFilter, CorpusRecord, type CreateScopeAnalystOptions, type CriuCapableClient, DefinePersonaInput, type DefinedLeaderboard, type DelegateOptions, type Deliverable, DeliverableSpec,
|
|
3997
|
+
export { Agent, AgentRunSpec, type AgentTurnBackend, type AgentTurnUsage, type AgenticOptions, type AgenticRunResult, type AgenticSurface, type AgenticTask, type AgenticTool, AnalystRegistry, type AnytimeReport, type AnytimeStrategySummary, type AnytimeTaskCurve, type ApplyContinuation, type ArtifactHandle, AssertTraceDerivedFindings, type AuditIntentInput, type AuditIntentOptions, type AuthorStrategyOptions, type AuthoredProfile, type AuthoredStrategy, type BenchmarkCell, type BenchmarkConfig, type BenchmarkLift, type BenchmarkReport, type BenchmarkStrategySummary, type BenchmarkTaskRow, Budget, type BudgetPool, type BudgetReadout, type ChampionPick, type ChampionPolicy, type CheckpointCapableBox, type CollectedAgentTurn, CombinatorShape, type CompletionAnalyst, type CompletionEvidence, type CompletionPolicy, type CompletionVerdict, CoordinationEvent, type CoordinationMcpHandle, Corpus, CorpusFilter, type CorpusReadbackOptions, CorpusRecord, type CreateScopeAnalystOptions, type CriuCapableClient, DefinePersonaInput, type DefinedLeaderboard, type DelegateOptions, type Deliverable, DeliverableSpec, Driver, type DriverAgentOptions, type DumbDriverOptions, type Environment, EqualKArm, EqualKOnCostOptions, EqualKVerdict, type EvolutionArchiveNode, type EvolutionAuthor, type EvolutionBandInfo, type EvolutionCandidate, type EvolutionGeneration, type EvolutionReport, ExecCtx, ExecutorConfig, ExecutorFactory, ExecutorRegistry, FanoutOptions, FanoutWinnerSelector, FileCorpus, type ForkCapableBox, type GitWorkspaceOptions, type HarvestCorpusOptions, type HarvestFailure, type HarvestReport, InMemoryCorpus, InMemoryResultBlobStore, type InMemoryRunContext, type InMemoryRunContextOptions, InMemorySpawnJournal, type InProcessOnPrompt, type InProcessPromptCtx, type InProcessSandboxClientOptions, type Inbox, type InboxMessage, type IntentAudit, type Interval, Iteration, type Leaderboard, type LeaderboardBenchScore, type LeaderboardBenchTask, type LeaderboardBenchmarkAdapter, type LeaderboardFlagSpec, type LeaderboardIterationInfo, type LeaderboardOptions, type LeaderboardRow, type LeaderboardRunContext, type LeaderboardScenario, type LeaderboardScore, type LeaderboardSpec, type LoopCampaignDispatchOptions, type LoopDispatchOptions, LoopLineageOptions, type LoopOptionsForDispatch, LoopResult, LoopShape, LoopTokenUsage, LoopUntilSpec, LoopWinner, MakeWorkerAgent, type McpEndpoint, type McpEnvironmentOptions, MountRecorder, type NaiveDriverOptions, type Observation, type ObserveInput, type ObserveOptions, type OpenSandboxRunBeforeStartContext, type OpenSandboxRunOptions, type OpenSandboxRunPromptOptions, Outcome, OutputAdapter, type PairwiseOptions, type PairwiseVerdict, PanelSpec, Persona, PipelineStage, type ProfileRichness, type ProfileRichnessThresholds, type PromotionGateOptions, type PromotionVerdict, type RegistryAnalyzeProjection, RenderCorpusToInstructionsOptions, type ReservationTicket, type ResolveSandboxClientOptions, ResultBlobStore, RouterConfig, type RunAgenticOptions, RunPersonifiedOptions, type SandboxCapabilities, SandboxClient, type SandboxLineage, type SandboxLineageHandle, type SandboxRun, SandboxRunAbortError, type SandboxToolPartState, Scope, ScopeAnalyst, ScopeAnalyzeInput, ScopeWidenGate, type SessionCapableBox, type SessionTraceBox, Settled, ShapeRegistry, type Shell, type ShotPersona, type ShotSpec, Spend, SteerContext, type SteeringDecision, type Strategy, type StrategyCtx, type StrategyEvolutionConfig, type StrategyResult, type StreamAgentTurnOptions, type SuperviseSurfaceOptions, type SuperviseSurfaceResult, SupervisedResult, Supervisor, SupervisorProfile, type SurfaceScore, type SurfaceWorkerConfig, type SurfaceWorkerOut, ToolLoopChat, ToolLoopCompactionOptions, type TraceSource, type TrajectoryAnalysis, TrajectoryReport, TrajectoryReportOptions, type TurnResult, UsageEvent, type UsageSink, Validator, type VerifierEnvironmentOptions, VerifySpec, type WatchTraceOptions, type WaterfallCollector, type WaterfallReport, type WaterfallSpan, WidenSpec, WinnerStrategy, type Workspace, type WorkspaceCommit, type WorkspaceRun, acquireSandbox, adaptiveRefine, analyzeTrace, anytimeReport, asAuthoredProfile, assertModelAllowed, assertStrategyContract, assertTraceDerivedFindings, assessAuthoredProfile, auditIntent, authorStrategy, authoredWorker, breadthStrategy, buildSteerContext, builtinShapes, collectAgentTurn, completionAuthorizes, contentAddress, createBudgetPool, createInMemoryRunContext, createInbox, createMcpEnvironment, createPushTraceSource, createSandboxLineage, createSandboxToolPartState, createScope, createScopeAnalyst, createShapeRegistry, createSupervisor, createVerifierEnvironment, createWaterfallCollector, decodeToolPart, defaultAnalystInstruction, defaultAuditorInstruction, defaultDelegateBudget, defaultProfileRichnessThresholds, defaultSelectWinner, defaultToolDetectors, defineLeaderboard, definePersona, defineStrategy, delegate, depthStrategy, deterministicCompletion, discriminatingMeans, driverAgent, dumbDriver, equalKOnCost, extractLlmCallEvent, failuresAnalyst, fanout, finalizeBestDelivered, flatWidenGate, gitWorkspace, harvestCorpus, inProcessSandboxClient, inlineSandboxClient, jjWorkspace, leaderboard, localShell, loopCampaignDispatch, loopDispatch, loopUntil, mapSandboxEvent, mapSandboxToolEvent, naiveDriver, observe, openSandboxRun, pairwiseSignificance, panel, pickChampion, pipeline, printBenchmarkReport, probeSandboxCapabilities, profileRichnessFinding, promotionGate, refine, registerShape, registryScopeAnalyst, renderAnytimeTable, renderCorpusToInstructions, renderLeaderboardHtml, renderLeaderboardMarkdown, renderLeaderboardSvg, renderPairwiseMarkdown, renderReport, reportLoopUsage, resolveSandboxClient, runAgentic, runBenchmark, runInWorkspace, runLoop, runPersonified, runStrategyEvolution, sample, sampleThenRefine, sandboxSessionTraceSource, selectChampion, selectValidWinner, sentinelCompletion, serveCoordinationMcp, settledToIteration, spendFromUsageEvents, stopSentinel, strategyAuthorContract, streamAgentTurn, sumSandboxUsage, superviseSurface, supervisorInstructions, trajectoryReport, verify, watchTrace, widen };
|
package/dist/loops.js
CHANGED
|
@@ -89,7 +89,7 @@ import {
|
|
|
89
89
|
watchTrace,
|
|
90
90
|
widen,
|
|
91
91
|
worktreeFanout
|
|
92
|
-
} from "./chunk-
|
|
92
|
+
} from "./chunk-4IBAMGBE.js";
|
|
93
93
|
import {
|
|
94
94
|
InMemoryResultBlobStore,
|
|
95
95
|
InMemorySpawnJournal,
|
|
@@ -131,7 +131,8 @@ import {
|
|
|
131
131
|
supervisorAgent,
|
|
132
132
|
supervisorInstructions,
|
|
133
133
|
workerFromBackend
|
|
134
|
-
} from "./chunk-
|
|
134
|
+
} from "./chunk-QSNSMJSZ.js";
|
|
135
|
+
import "./chunk-DPEUKJRO.js";
|
|
135
136
|
import "./chunk-UD4BHQMI.js";
|
|
136
137
|
import {
|
|
137
138
|
createAgentEnvironmentProviderRegistry,
|
|
@@ -149,7 +150,6 @@ import {
|
|
|
149
150
|
} from "./chunk-FVJ7M3DA.js";
|
|
150
151
|
import "./chunk-7LO5GMAO.js";
|
|
151
152
|
import "./chunk-YEJR7IXO.js";
|
|
152
|
-
import "./chunk-DPEUKJRO.js";
|
|
153
153
|
import "./chunk-DGUM43GV.js";
|
|
154
154
|
export {
|
|
155
155
|
FileCorpus,
|
package/dist/mcp/bin.js
CHANGED
|
@@ -9,13 +9,13 @@ import {
|
|
|
9
9
|
DelegationTaskQueue,
|
|
10
10
|
FileDelegationStore,
|
|
11
11
|
createMcpServer
|
|
12
|
-
} from "../chunk-
|
|
12
|
+
} from "../chunk-QSNSMJSZ.js";
|
|
13
|
+
import "../chunk-DPEUKJRO.js";
|
|
13
14
|
import "../chunk-UD4BHQMI.js";
|
|
14
15
|
import "../chunk-BZF3KQ6G.js";
|
|
15
16
|
import "../chunk-FVJ7M3DA.js";
|
|
16
17
|
import "../chunk-7LO5GMAO.js";
|
|
17
18
|
import "../chunk-YEJR7IXO.js";
|
|
18
|
-
import "../chunk-DPEUKJRO.js";
|
|
19
19
|
import "../chunk-DGUM43GV.js";
|
|
20
20
|
|
|
21
21
|
// src/mcp/delegate-supervisor-provisioning.ts
|
package/dist/mcp/index.d.ts
CHANGED
|
@@ -1,19 +1,18 @@
|
|
|
1
|
-
import { S as SandboxClient, d as LoopSandboxPlacement } from '../types-
|
|
2
|
-
import { F as FleetHandle, D as DelegationExecutor, a as DelegationTaskQueue, b as FeedbackStore, c as DelegateFeedbackResult, d as DelegateFeedbackArgs, U as UiAuditorDelegate, e as DelegateUiAuditArgs, f as DelegateUiAuditResult, g as DelegationHistoryResult, h as DelegationHistoryArgs, i as DelegationStatusResult, j as DelegationStatusArgs } from '../coordination-
|
|
3
|
-
export { A as AnalystRegistry, C as CappedDelegationTrace, k as CoderDelegate, l as CoderReview, m as CoderReviewer, n as CoordinationEvent, o as CoordinationTools, p as CoordinationToolsOptions, q as DELEGATE_DESCRIPTION, r as DELEGATE_INPUT_SCHEMA, s as DELEGATE_TOOL_NAME, t as DELEGATION_TRACE_MAX_BYTES, u as DELEGATION_TRACE_MAX_SPANS, v as DelegateArgs, w as DelegateCodeArgs, x as DelegateCodeConfig, y as DelegateCodeResult, z as DelegateHandlerOptions, B as DelegateResearchArgs,
|
|
1
|
+
import { S as SandboxClient, d as LoopSandboxPlacement } from '../types-Dnk189QA.js';
|
|
2
|
+
import { F as FleetHandle, D as DelegationExecutor, a as DelegationTaskQueue, b as FeedbackStore, c as DelegateFeedbackResult, d as DelegateFeedbackArgs, U as UiAuditorDelegate, e as DelegateUiAuditArgs, f as DelegateUiAuditResult, g as DelegationHistoryResult, h as DelegationHistoryArgs, i as DelegationStatusResult, j as DelegationStatusArgs } from '../coordination-DxHduZg7.js';
|
|
3
|
+
export { A as AnalystRegistry, C as CappedDelegationTrace, k as CoderDelegate, l as CoderReview, m as CoderReviewer, n as CoordinationEvent, o as CoordinationTools, p as CoordinationToolsOptions, q as DELEGATE_DESCRIPTION, r as DELEGATE_INPUT_SCHEMA, s as DELEGATE_TOOL_NAME, t as DELEGATION_TRACE_MAX_BYTES, u as DELEGATION_TRACE_MAX_SPANS, v as DelegateArgs, w as DelegateCodeArgs, x as DelegateCodeConfig, y as DelegateCodeResult, z as DelegateHandlerOptions, B as DelegateResearchArgs, G as DelegateResearchConfig, H as DelegateResearchResult, I as DelegateResult, J as DelegateRunCtx, K as DelegateUiAuditConfig, L as DelegateUiAuditRoute, N as DelegationError, O as DelegationFeedbackSnapshot, P as DelegationHistoryEntry, Q as DelegationPersistenceError, R as DelegationProfile, S as DelegationProgress, T as DelegationRecord, V as DelegationResultPayload, W as DelegationResumeContext, X as DelegationResumeDriver, Y as DelegationResumeTick, Z as DelegationRunContext, _ as DelegationStateCorruptError, $ as DelegationStatus, a0 as DelegationStore, a1 as DelegationTaskQueueOptions, a2 as DelegationTraceCaps, a3 as DelegationTraceCollector, a4 as DelegationTraceSpan, a5 as DetachedSessionDelegateOptions, a6 as DetachedSessionRefParts, a7 as DetachedTurn, a8 as DetachedTurnResumeDriverOptions, a9 as DetachedWinnerSelection, aa as DriveTurnCapableBox, ab as DriveTurnTick, ac as FeedbackEvent, ad as FeedbackRating, ae as FeedbackRefersTo, af as FileDelegationStore, ag as FileDelegationStoreOptions, ah as FleetWorkspaceExecutorOptions, ai as InMemoryDelegationStore, aj as InMemoryFeedbackStore, ak as JsonRpcMessage, al as JsonRpcResponse, M as MakeWorkerAgent, am as McpServer, an as McpServerOptions, ao as McpToolDescriptor, ap as McpTransport, aq as Question, ar as QuestionDecision, as as QuestionPolicy, at as QuestionRecord, au as ResearchOutputShape, av as ResearchSource, aw as RunDetachedTurnOptions, ax as SettleDetachedCoderTurnOptions, ay as SettledWorker, az as SiblingSandboxExecutorOptions, aA as SubmitInput, aB as SubmitOutput, aC as TraceContext, aD as UiAuditorDelegationOutput, aE as buildDelegationTraceSpans, aF as capDelegationTrace, aG as coderTaskFromArgs, aH as composeLoopTraceEmitters, aI as createCoordinationTools, aJ as createDelegateHandler, aK as createDelegationTraceCollector, aL as createDetachedTurnResumeDriver, aM as createFleetWorkspaceExecutor, aN as createInProcessTransport, aO as createMcpServer, aP as createPropagatingTraceEmitter, aQ as createSiblingSandboxExecutor, aR as detachedSessionDelegate, aS as detachedTurnEvents, aT as eventToSnapshot, aU as formatDetachedSessionRef, aV as hashIdempotencyInput, aW as parseDetachedSessionRef, aX as readTraceContextFromEnv, aY as runDetachedTurn, aZ as settleDetachedCoderTurn, a_ as traceContextToEnv, a$ as validateDelegateArgs } from '../coordination-DxHduZg7.js';
|
|
4
4
|
import { L as LocalHarness, r as runLocalHarness } from '../local-harness-sI0S_XNA.js';
|
|
5
5
|
export { a as LocalHarnessResult, R as RunLocalHarnessOptions } from '../local-harness-sI0S_XNA.js';
|
|
6
|
-
import { G as GitRunner } from '../
|
|
7
|
-
export { C as CreateWorktreeOptions, b as DiffOptions, c as DiffResult, R as RemoveWorktreeOptions, d as WorktreeHandle, e as captureWorktreeDiff, f as createWorktree, r as removeWorktree } from '../
|
|
6
|
+
import { G as GitRunner } from '../completion-gate-BDaBP8dd.js';
|
|
7
|
+
export { C as CreateWorktreeOptions, b as DiffOptions, c as DiffResult, R as RemoveWorktreeOptions, d as WorktreeHandle, e as captureWorktreeDiff, f as createWorktree, r as removeWorktree } from '../completion-gate-BDaBP8dd.js';
|
|
8
8
|
export { C as CreateKbGateOptions, F as FactCandidate, a as FactJudge, b as FactJudgeVerdict, K as KbGateResult, c as createKbGate } from '../kb-gate-CwHO0vz6.js';
|
|
9
|
-
export { m as mcpToolsForRuntimeMcp, a as mcpToolsForRuntimeMcpSubset } from '../openai-tools-
|
|
9
|
+
export { m as mcpToolsForRuntimeMcp, a as mcpToolsForRuntimeMcpSubset } from '../openai-tools-D0xmzo0y.js';
|
|
10
10
|
import { AnalystFinding } from '@tangle-network/agent-eval';
|
|
11
11
|
import { U as UiLens } from '../substrate-DO2GHNg2.js';
|
|
12
12
|
import '@tangle-network/sandbox';
|
|
13
|
-
import '../types-
|
|
13
|
+
import '../types-DAJQRIUD.js';
|
|
14
14
|
import '@tangle-network/agent-interface';
|
|
15
|
-
import '../
|
|
16
|
-
import '../router-client-r8y_VFVM.js';
|
|
15
|
+
import '../router-client-DJImUDlm.js';
|
|
17
16
|
import '../environment-provider.js';
|
|
18
17
|
import '@tangle-network/agent-interface/environment-provider';
|
|
19
18
|
import 'node:child_process';
|